Use per-core mutex for XSCOM for P8 errata.

    See HW822317.  The HMER register in P8 is not implemented to
    handle multi-threaded XSCOM properly, so we need to move the
    XSCOM mutex from per-thread to per-core.  Also, there is an
    issue where the 'done' bit can come on 1 cycle before the error
    indicators, so need to potentially read the HMER a second time.

Change-Id: I495031a6e425fe7d5c6ffef8dda1e7a71caac9f2
CQ: SW250902
Backport: release-fips810
Reviewed-on: http://gfw160.aus.stglabs.ibm.com:8080/gerrit/9929
Reviewed-by: Michael Baiocchi <baiocchi@us.ibm.com>
Reviewed-by: A. Patrick Williams III <iawillia@us.ibm.com>
Tested-by: Jenkins Server

1 files changed, 2 insertions, 2 deletions

diff --git a/src/include/kernel/cpu.H b/src/include/kernel/cpu.H
index 824a65370..478983148 100644
--- a/src/include/kernel/cpu.H
+++ b/src/include/kernel/cpu.H
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* IBM CONFIDENTIAL                                                       */
 /*                                                                        */
-/* COPYRIGHT International Business Machines Corp. 2010,2012              */
+/* COPYRIGHT International Business Machines Corp. 2010,2014              */
 /*                                                                        */
 /* p1                                                                     */
 /*                                                                        */
@@ -89,7 +89,7 @@ struct cpu_t
     task_t* idle_task;
 
     /** XSCOM mutex to serialize access per CPU */
-    mutex_t xscom_mutex;
+    mutex_t* xscom_mutex;
 
     /** counter for executePeriodics */
     size_t periodic_count;